International Journal of Computer Applications (0975 – 8887)

Volume 181 – No. 49, April 2019

Computational Framework for Heart Disease Prediction

using Deep Belief Neural Network with Fuzzy Logic

S. Saranya R. Manavalan, PhD

Research Scholar, Assistant Professor in Computer Science, o/o
Department of Computer Science, Head, Department of Computer Applications &
Arignar Anna Government Arts College, Villupuram, Information Technology, Arignar Anna Government
Tamil Nadu, India Arts College, Villupuram, Tamil Nadu, India

ABSTRACT (WHO) has estimated that 12 million deaths occur worldwide,

Cardiovascular Disease (CVD) is the main problems of every year death rate is increased due to the Heart diseases. In
morbidity and mortality in the current lifestyle. The prediction 2008, 17.3 million people died due to Heart Disease. Over
of heart disease is the most complicated task in the field of 80% of deaths in world are caused only the reason of Heart
medical sciences. Data mining techniques have been widely disease. In 2016, Cardiovascular disease (CVDs) take the
applied in bioinformatics to analyze biomedical data. Today, lives of 17.9 million people every year, 31% of all global
the field of medicine has come a long way to treat patients deaths. WHO estimated by 2030, almost 23.6 million people
with various types of diseases. Heart disease is one of the will die due to Heart disease [1]. Data Mining techniques is
most threatening diseases that cannot be seen with the naked most efficient techniques used for predicting the heart
eye. Poor and incorrect clinical decisions result in the death of diseases. The mining of valuable information for diagnosis is
a patient who cannot be authorized. Today the health sector not an easy task, since the complexity and toughness of
contains a large amount of health data, which contain hidden information in medical domain is too high.
information. Advanced data extraction techniques and Several risk factors for heart disease include i) family history
computer-generated information are used for predicting heart of heart disease, ii). smoking, iii) cholesterol, iv) high blood
disease at an earlier stage. In this paper, classification pressure, v) obesity, vi) diabetes, vii) stress and viii) lack of
techniques such as Deep Belief Network (DBN) and Fuzzy physical exercise. Heart disease is a broad term that includes
Deep Belief Network (FDBN) are proposed and analyzed on all types of diseases affecting different components of the
the cardiovascular disease dataset. The performance of these heart.
techniques is evaluated using evaluation parameter such as
accuracy and error rate. From the experimental results, it is Some types of heart diseases are i) Coronary Heart Disease,
proved that the accuracy of the Fuzzy Deep Belief network is ii) Angina Pectoris, iii) Congestive Heart Failure, iv)
superior to the Deep Belief network. Further the analysis of Cardiomyopathy, v) Congenital Heart Disease, vi)
the experiment also showed that the FDBN system is one of Arrhythmias and vii) Myocarditis.
the best classification models in efficiently predicting
cardiovascular diseases with the lowest error rate and 1.1 Symptoms of a Heart Attack
maximum accuracy. The signs and symptoms of Coronary Heart Disease (CHD)
may differ between women and men. Some women who have
Keywords CHD without any signs or symptoms. This is called silent
Heart Disease Prediction System, Deep Belief Network, CHD. Silent CHD may not be diagnosed until a woman has
Fuzzy Deep Belief Network classification algorithm. signs and symptoms of a heart attack, heart failure, or an
Arrhythmia (irregular heartbeat). Symptoms of a heart attack
1. INTRODUCTION are given below:
Data Mining is an essential process where intelligent methods
 Discomfort, pressure, heaviness, or pain in the chest,
are required to extract valuable data patterns from huge data
arm, or below the breastbone.
set. Heart Diseases remain the biggest cause of deaths for the
last two decades. Recently computer technology and machine  Discomfort radiating to the back, jaw, throat, or arm.
learning techniques are used to develop software to assist
doctors in making decision of heart disease at the early stage.  Fullness, indigestion, or choking feeling (may feel like
The diagnosis of heart disease is highly depends on clinical heartburn).
and pathological data. Heart disease prediction system can
 Sweating, nausea, vomiting, or dizziness.
assist medical professionals in predicting heart disease status
based on the clinical data of the patients. In biomedical field,  Extreme weakness, anxiety, or shortness of breath.
data mining plays an essential role for prediction of diseases
and staging the same.  Rapid or irregular heartbeats.

In biomedical diagnosis, the information provided by the This paper is organized as follows: section 2 describes
patients may include redundant and interrelated symptoms Material and Methods, section 3 presents the proposed
and signs especially when the patients suffer from more than method for Heart Disease Prediction, Experimental results and
one type of disease of the same category. The health care its analysis are comprehensively discussed in section 4 and
industries are facing many challenges and issues based on the section 5 concluded with wrapping up of the current work.
patient’s severity which is to be reduced and detect it earlier
in a more effective way. The World Health Organization

26
 International Journal of Computer Applications (0975 – 8887)
Volume 181 – No. 49, April 2019

2. MATERIAL AND METHODS KNN yielded the accuracy between 94% and 97.4% for
The ability to extract useful hidden knowledge in the large different values of K. When K = 7, the method achieved the
amount of data and to act on the same is a challenging task, maximum accuracy and specificity are 97.4% and 99%
since growth of the data size is enormously increasing day by respectively.
day. The need to understand large, complex, enriched In 2012, Nidhi et al. [7] introduced the algorithms Naïve
information in data sets has now increased in all the varied Bayes, Decision Tree and Neural Network and the same was
fields of technology, business, medical and science. The analyzed. The analysis results showed that the neural network
process of developing a novel Computer Based Information produced at maximum accuracy for 15 attributes i.e. nearly
System (CBIS) for discovering potentially useful knowledge 100%. On the other hand, Decision Tree has also performed
from data is necessary to increase the life time of the patient. well for the same number of attributes and yielded 99.62% of
The various data mining techniques introduced so for to heart accuracy. Moreover, Genetic Algorithm and Feature Subset
disease predictions are reviewed extensively here under: Selection methods are used to reduce the number of attributes
In 2010, Anbarasi et al. [2] proposed enhanced prediction to six and given as input to classification models. One of the
model for Heart Disease through the hybridization of genetic classification model Decision Tree produced almost same
algorithm with Decision Tree for feature subset selection, accuracy.
Genetic with Naive Bayes and Genetic with classification via
clustering. The enhanced prediction of heart disease is carried In 2012, Pethalakshmi et al. [8] introduced the algorithm such
out using genetic algorithm based feature subset selection as Fuzzy Decision tree, Fuzzy Naive Bayes, Fuzzy Neural
method with 10 attributes. The prediction accuracy is Network, Fuzzy k-means to predict the heart disease. The
increased while incorporating classification via clustering accuracy of these algorithms were 90.06%, 89.62%, 91.09%
with feature subset selection algorithms. Classification and 99.49% respectively. Fuzzy k-means yielded the best
techniques such as Naïve Bayes, Decision Tree and result which was at least 8.4% more than others.
Classification by clustering were used for prediction. In 2013, Abishek [9] proposed J48 and Naive Bayes to predict
the Heart disease and adopted Weka tool to implement the
Kumari et al. [3] applied data mining techniques such as above techniques. Accuracy of these algorithms is 95.56%
SVM, ANN, Decision Tree, RIPPER classifier to predict the and 92.42% respectively. The accuracy of j48 was 3.14%
risk of heart diseases, in 2011. The performance of these higher than Naive Bayes algorithm.
algorithms is analyzed using statistical analysis factors such as
Accuracy and Error Rate. Accuracy of RIPPER, Decision In 2013, Chitra et al. [10] implemented the algorithm such as
Tree, ANN and SVM are 81.08%, 79.05%, 80.06% and Artificial Neural Network (ANN), K-Means Clustering and
84.12% respectively whereas error rates of RIPPER, Decision Fuzzy C Means Clustering to predict the heart disease. The
Tree, ANN and SVM are 2.756, 0.2755, 0.2248 and 0.1588 accuracy of these algorithms were 85%, 88% and 92%
respectively. The analysis clearly showed that out of these respectively. Fuzzy C Means Clustering yielded the best result
four classification models, SVM predicted cardiovascular which is at least 4% more than others.
disease with least error rate and highest accuracy. In 2013, Patel et al. [11] applied the algorithm such as
Sundar, et al. [4] discussed the performance analysis of Decision Tree and Naive Bayes to predict the heart disease.
classification techniques Naive Bayes and Weighted The accuracy of these algorithms were 99.2%, 96.5% and
Association Classifier (WAC) for heart disease prediction 88.3% respectively. Decision Tree yielded the best result
based on evaluation using classification matrix, which revels which is at least 2.7% more than others.
the frequency of correct and incorrect prediction. Accuracy of In 2013, Vikas [12] applied CART Classification to predict
WAC and Naive Bayes are 84% and 78% respectively. The the heart disease. Accuracy of the algorithm is 84.49% and the
accuracy of WAC is 6% higher than Naive Bayes. time consumed by the algorithm is 0.23s.
In 2012, Dangare et al. [1] used 13 medical related input In 2014, Waghulde et al. [13] proposed the models called as
attributes medical terms such as sex, blood pressure, and Neural Network and Genetic Algorithm for the prediction of
cholesterol etc for prediction. To get more appropriate results, heart disease. The accuracy of the system is 98% respectively.
additionally added two more important attributes with above
attributes i.e. obesity and smoking for heart disease. In 2014, Niranjana Devi et al. performed a research work,
Multilayer Perceptron Neural Network (MLPNN), Back Evolutionary-Fuzzy Expert System for the Diagnosis of
propagation algorithm is introduced for heart disease. The Coronary Artery Disease. In this, a fuzzy expert system with
prediction method yielded the accuracy of 99.25% for 13 Genetic Algorithm is proposed to diagnose CAD disease
attributes and nearly 100% for 15 attributes. condition. Genetic Algorithm is used to optimize the
membership function parameters. The proposed system is
In 2012, Peter et al. [5] introduced the algorithm such as validated over CAD dataset and achieved an accuracy of
Naive Bayes, Multilayer, J48 and KNN and conducted 88.79%.
experimentation on dataset of health care domain. The
Correlation based Feature Selection and filter subset Rupali et al. [15] proposed the Classification based algorithms
evaluation methods are adapted to reduce more number of Naive Bayes and Laplace smoothing are introduced to predict
irrelevant and redundant attributes thereby increases the the heart disease in 2014. The accuracy of these algorithms
performance of classifiers. The accuracy of these algorithms were 78% and 86% respectively. Classification using Laplace
are 85.18%, 78.88%, 88.18% and 85.55% respectively and Smoothing yielded the best result than others.
provide the evidence of significant improvement in the
results. Venkatalakshmi et al. [16] introduced the algorithm such as
Naive Bayes and Decision Tree to predict the heart disease in
In 2012, Shouman et al. [6] proposed the model called as K- 2014. The accuracy of these algorithms is 85.03% and 84.01%
Nearest Neighbour for the prediction of heart disease. In this respectively. Naive Bayes yielded the higher result which is at
model, the value of K is ranged between one and thirteen. least 1.02% more than others.

27
 International Journal of Computer Applications (0975 – 8887)
Volume 181 – No. 49, April 2019

In 2015, D’Souza [17] applied three techniques such as From this literature survey, the following issues are identified
Artificial Neural Network, K Means Clustering and Apriori in heart disease prediction using computational models in
Algorithm to classify whether the patients have the heart health care industry data.
disease or not and their performance are analysed. The results
showed that Artificial Neural Networks outperform well  Prediction accuracy is low with reduced number of
compare to others. attributes.

In 2015, Adbar et al. [18] proposed methods C5.0, Neural  It take more time for predict.
Network, SVM, and KNN to predict the risk of heart diseases.  More false classification.
Accuracy of c5.0, Neural Network, SVM, and KNN are
93.02%, 89.4%, 86.05% and 80.23% respectively. C5.0  The choice of the kernel and other parameter
performed well when compared to others. Further, it was also selections.
noted that the accuracy of C5.0 is at least 3.62% higher than
others.  The limit on ability to categorize correctly.

In 2016, Patel et al. [19] applied methods J48 and LMT  Poor clinical decisions lead to mortality.
algorithm to predict the heart diseases. Performances of these  The noise and missing vales make a hurdle to design
algorithms are analyzed in terms of accuracy and time classification model.
complexity. The accuracy of J48 and LMT are 56.76% and
55.77% respectively whereas time consumption these  The irrelevant or redundant attributes Removal is
algorithms are 0.04s and 0.39s respectively. difficult task.
In 2016, Rajalakshmi et al. [20] proposed Weighted  Accuracy and Speed of the model only for some
Association Classifier (WAC) and K-Means Clustering to extents.
predict the heart disease. Accuracy of these algorithms is
93.89% and 92.84% respectively. Combination of these  The time to construct the model (training time) is
algorithms produced maximum accuracy of 94.54%. high.

In 2016, Suganya et al. [21] proposed CART classifier to  Lack of handling noise and missing values
predict the heart disease. Minimum distance CART classifier
is used to classify the data among various groups. The  Interpretability i.e. Level of understanding and insight
accuracy of the CART classifier algorithm is 83%. provided by the model.

In 2018, Wadhawan [22] applied combination of K-Means 3. PROPOSED METHOD

and Apriori Algorithm for heart diseases prediction. The
method incurred the accuracy of 74%. The Apriori technique 3.1 Computational Model for Heart
found strong rules for prediction. The proposed method Disease Prediction
yielded results as good as a neural network. The proposed model for Heart Disease Prediction System is
developed using Deep Learning approach with fuzzy logic.
2.1 Issues in Prediction of Heart Disease The Figure 1 is presenting the overall work flow of the
Using Computational Method proposed system.

Figure 1. Computational Model for Heart Disease Prediction

3.1.1 Deep Belief Network

A DBN is a deep learning technique that learns by composing
multiple RBM layers. The RBM, which is based on the
Hopfield network, uses the energy function and obtains unit
values in a probabilistic way (using a Boltzmann distribution).
The RBM consists of a visible unit layer and a hidden unit
layer.

Figure 2. The Structure of RBM

In the structure of RBM, the hidden unit layer acts as the
previous visible unit layer. DBN learning is done by
configuring the visible layer and hidden layer 1 into a single
RBM. Once learning is complete, hidden layers 1 and 2 are

28
 International Journal of Computer Applications (0975 – 8887)
Volume 181 – No. 49, April 2019

trained via the RBM by giving a new input as a value of For fuzzy sets, membership function takes values in the
hidden layer 1. As such, learning is sequential up to the last interval [0, 1]. The interval between 0 and 1 is called
layer. membership level or degree of membership.
A supervised learning-based classification technique using the
DBN is the back propagation algorithm, which is configured A fuzzy set A is defined as:
in the uppermost layer in the DBN. A classification prediction ∈ ∈ (2)
model using the neural network backpropagation-DBN was
implemented for heart disease prediction. where is a membership function belongs to the interval
[0, 1].
Initially, the DBN built a learning model using a training set.
Eight input variables were used (age, sex, chest_pain_type,
Trestbps, chol, Fbs, Thalach, Ca) and 1 output data. The DBN
3.1.2.1 Membership Functions
Membership function describes the composition of the
consisted of two steps. The first phase was the construction of
the RBM network work using unsupervised learning. The basic elements of the series x in the fuzzy set X, so can
RBM parameters such as epoch, batchsize and momentum are assume a large class of functions. Reasonable functions are
set as 1, 10 and 0.5 respectively. In the second phase, the often linear functions for parts, such as triangular or
RBM network learned the neural network backward trapezoidal functions.
propagation algorithm of supervised learning. Triangular Membership Function: Let a, b, c x be the
Table1. Algorithm- Deep Belief Network coordinates of the three vertices of in a fuzzy set A (a:
Lower limit c: upper limit, whose degree of belonging is zero,
Deep Belief Network Algorithm for Classify b: center where the degree of belonging is 1).
Heart Disease Data set
Input: Heart Disease Dataset
Output: Classified Heart Disease Dataset
Step 1. Collect the heart disease Dataset from
Statlog heart disease database
0 if x ≤ a
Step 2. Separate into Training and Test Sets using
k-fold cross validation
Step 3. Define a Deep Belief Network Structure A(X) =
Step 4. Set parameters, values, initialize weights
Step 5. Transform data to network inputs 0 if x ≥ c (3)
Step 6. Start DBN training using Greedy layer-
wise deep training
Step 7. Construct an RBM with an input layer v
and a hidden layer h, Train the RBM
Step 8. Stop and Test
Step 9. Implementation: use the Deep Belief
Network with new cases

3.1.2 Fuzzy Set

The fuzzy sets are derived from the concept of set theory.
Fuzzy sets can be considered an extension of the classic sets.
In a classic set (or light set), the set elements are called
elements or members of the set. An element x belonging to a Figure 3. Triangular Membership Function
set A is defined as X ∈ A, an element that is not a member A
is referred to as x ∉ A. A function feature function or Table 2. Algorithm- Fuzzy Deep Belief Network
membership is defined as an element in the universe U
has a crisp value of 1 or 0. For every x ∈ U. Deep Belief Network Algorithm for Classify
Heart Disease Dataset
Input: Heart Disease Dataset
Output: Classified Heart Disease Dataset
1 for x∈ A Step 1. Collect the heart disease Dataset from
A(X) = Statlog heart disease database
0 for x ∈ A (1) Step 2. Fuzzy Membership function is applied to
This also can be expressed as ∈ remove the uncertainty in the data
Step 3. Separate into Training and Test Sets using

29
 International Journal of Computer Applications (0975 – 8887)
Volume 181 – No. 49, April 2019

k-fold cross validation 4.2 Cross-Validation

Step 4. Define a Deep Belief Network Structure Cross-Validation is one of the statistical methods used for
evaluating the performance of learning algorithms by dividing
Step 5. Set parameters, values, initialize weights input data into two segments: one used to train a model and
the other used to validate the model. In typical cross-
Step 6. Transform data to network inputs
validation, the training and validation sets must cross-over in
Step 7. Start DBN training using Greedy layer- successive rounds such that each data point has a chance of
wise deep training being validated against. The basic form of the cross-validation
is k-fold cross-validation.
Step 8. Construct an RBM with an input layer v
and a hidden layer h, Train the In k-fold cross-validation the data is first partitioned into k
equally (or nearly equally) sized segments or folds.
Step 9. Stop and Test Subsequently, k iterations of training and validation are
Step 10. Implementation: use the Deep Belief performed such that each iteration a different fold of the data
Network with new cases is held-out for validation while the remaining k-1 folds are
used for learning. In data mining and machine learning 10-
fold cross-validation (k= 10) is the most common for the
evaluation of classification algorithms.
4. EXPERIMENTAL RESULT AND
ANALYSIS Cross-validation is used to evaluate or compare learning
algorithms as follows: in each iteration, one or more learning
4.1 Data Source algorithms use k-1 folds of data to learn one or more models,
A total (factors) were obtained from UCI/heart-statlog.arff. and subsequently, the learned models are asked to make
The dataset consists of 13 attributes and 270 records. predictions about the data in the validation fold. The
However, only 9 attributes are used for this research. performance of each learning algorithm on each fold can be
The attribute “Diagnosis” was identified as the predictable tracked using some pre-determined performance metric like
attribute with value “1” for patients with heart disease and accuracy.
value “0” for patient with no heart disease. The key attribute
and their type with description is shown in table 3.
4.3 Evaluation of Classification Algorithm:
Accuracy: The accuracy of a test is the ability to differentiate
Table 3. Selected Attribute for Heart Disease Prediction the patient and healthy cases correctly. To estimate the
accuracy of a test, the proportion of true positive and true
Name Type Description negative is compared for all evaluated cases. Mathematically,
Age Continuous Age in years this can be stated as:

Sex Discrete 1=male

0=female
The Misclassification Rate can be stated as
Cp Discrete Chest Pain Type
1=Typical angina,
2=Atypical angina,
3=Non-anginal, True Positive (TP) = the number of cases correctly identified
4=Asymptomatic as patient

Trestbps Continuous Resting Blood Pressure (in False Positive (FP) = the number of cases incorrectly
mm Hg) identified as patient

Chol Continuous Serum Cholesterol in mg/dl True Negative (TN) = the number of cases correctly
identified as healthy
Fbs Discrete Fasting Blood Sugar > 120
mg/dl False Negative (FN) = the number of cases incorrectly
identified as healthy
Thalach Continuous Maximum heart rate
achieved 4.4 Result and Discussion
This section presents the experimental results and analysis
Ca Continuous Number of major vessels done for this study. The research work, two classification
colored by fluoroscopy that models such as DBN and Fuzzy DBN were evaluated. Using
ranged between 0 and 3. 10 fold validation to access the performance of them. The
Diagnosis Discrete Diagnosis classes proposed algorithms are implemented using MATLAB.

0- healthy, 1- possible The statistical parameters accuracy is used as measure to

heart disease access the ability of the prediction algorithms. Performance of
DBN and Fuzzy DBN classifier is evaluated by 10 fold
validation and their outcome is presented in table 4
respectively.

30
 International Journal of Computer Applications (0975 – 8887)
Volume 181 – No. 49, April 2019

Table 4. Accuracy of 10-fold validation 9 0.7037 0.8519

METHOD
10 0.6667 0.9259
FOLD Deep Belief Fuzzy Deep Belief
Network Network
From the experimental result, it was observed that the Fuzzy
1 0.6667 0.963
DBN method produces high accuracy values for the entire
2 0.8148 0.8889 folds while compared to DBN. The maximum accuracy
achieved by Fuzzy DBN is 96.3% for fold 1 and 8, where as
3 0.7407 0.8519 DBN achieved a maximum accuracy of 81.48% for the 2nd
fold only.
4 0.7407 0.8889
The graphical representations of accuracy produced by the
5 0.7037 0.9259 methods are presented in figure 4. The x-axis represents
number of folds and y-axis represents accuracy in terms of
6 0.7037 0.9259 percentage. It is observed that the Fuzzy DBN classifier
achieved high performance than DBN. It is clearly seen that
7 0.6667 0.8889 the Fuzzy DBN achieved the highest accuracy of 93.6% and
the lowest accuracy of 85.19%, while the maximum and
8 0.7778 0.963 minimum accuracy achieved by DBN is 81.48% and 66.6%
respectively.

1.2

0.8
Deep Belief Network
0.6
Fuzzy Deep Belief
0.4 Network

0.2

0
1 2 3 4 5 6 7 8 9 10

Figure 4. Accuracy of for DBN and FDBN for 10 folds

The error rate for Deep Belief Network and Fuzzy Deep 6 0.2963 0.0741
Belief Network for the 10 folds are shown in table 5.
Table 5. Error Rate of 10-fold validation 7 0.3333 0.1111

METHOD 8 0.2222 0.037

FOLD Fuzzy Deep
Deep Belief 9 0.2963 0.1481
Belief
Network
Network
10 0.3333 0.0741
1 0.3333 0.037
Table 5 shows that Fuzzy DBN method produces low error
2 0.1852 0.1111 values for all folds when compared to DBN. Fuzzy DBN heart
disease prediction technique achieved lowest error value of
3 0.2593 0.1481 0.037 for Fold 1 and Fold 8.
The graphical representation of error value produced by the
4 0.2593 0.1111 methods while evaluate the heart disease dataset using DBN
and FDBN is presented in figure 5.
5 0.2963 0.0741

31
 International Journal of Computer Applications (0975 – 8887)
Volume 181 – No. 49, April 2019

0.35
0.3
0.25
0.2
DBN-ERR
0.15
FDBN-ERR
0.1
0.05
0
1 2 3 4 5 6 7 8 9 10

Figure 5. Error Rate Comparison for DBN and FDBN

The average computational results for ten fold validation Table 6. Classifier Performance Measurement Accuracy
presented in table 6 and their graphical representation is also
exposed in figure 6. Fuzzy Deep
Deep Belief
Methods Belief
Network
Network
Average
0.7185 0.9074
Accuracy

AVERAGE ACCURACY
1

0.8

0.6

0.4 ACCURACY
0.2

0
DBN FDBN

Figure 6. Average Accuracy of DBN and FDBN

The proposed Fuzzy DBN Classifier is always producing Table 7. Classifier Performance Measurement - Error
better results than DBN. The Fuzzy DBN yields the average
accuracy value of 0.9074 which is 18.89% higher than DBN. Fuzzy Deep
Deep Belief
Methods Belief
Network
The average error for ten fold validation of the both methods Network
are presented in table 7 and their graphical representation is
also exposed in figure 7. Average
0.2814 0.0925
Error

In Table 7, the average error value of FDBN is 0.0925 which

is 0.19% lesser than the error value achieved by DBN.

32
 International Journal of Computer Applications (0975 – 8887)
Volume 181 – No. 49, April 2019

AVERAGE ERROR
0.3
0.25
0.2
0.15
ERROR
0.1
0.05
0
DBN FDBN

Figure 7. Average Error Measure value of DBN and FDBN

[4] N. Aditya Sundar, P. Pushpa Latha, M. Rama Chandra,
5. CONCLUSION “Performance Analysis of Classification Data Mining
The main purpose of this paper is to provide an insight about
Techniques over Heart Disease Data Base”, International
early detection of heart disease using data mining techniques.
Journal of Engineering Science and Advanced
The prediction of Heart disease is implemented through Deep
Technology, Vol. 2, No. 3, pp. 470-478, 2013.
Belief Network and Fuzzy Deep Belief Network. The
proposed models can assist medical professionals in [5] T. John Peter and K. Somasundaram, “Study and
evaluating a patient’s heart disease based on the clinical data Development of Novel Feature Selection Framework for
of the patient. The proposed research work is evaluated using Heart Disease Prediction”, International Journal of
9 essential medical attributes such as Age, Sex, Chest Pain Scientific and Research Publications, Vol. 2, No. 10, pp.
Type, Resting Blood Pressure, Serum Cholesterol, Fasting 1- 7, 2012.
Blood Sugar, Maximum heart rate and the number of major
vessels colored by fluoroscopy that ranged between 0 and 3. [6] Mai Shouman, Tim Turner and Rob Stocker “Applying
The performance of the DBN is compared over fuzzy DBN. K-Nearest Neighbour in Diagnosing Heart Disease
Deep Belief Network and Fuzzy Deep Belief Networks were Patients”, Proceedings of International Conference on
compared on the basis of Accuracy and Error-Rate. Fuzzy Knowledge Discovery, pp. 23-29, 2012.
DBN achieved an accuracy of 90.74%, which is 18.89% [7] Nidhi Bhatla and Kiran Jyoti, “An Analysis of Heart
higher than the Deep belief Network. The average error rate of Disease Prediction using Different Data Mining
Fuzzy DBN is 9.25%, which is 18.89% lower than the error Techniques”, International Journal of Engineering
rate achieved by DBN. The experimental result clearly proved Research and Technology, Vol. 1, No. 8, pp. 1-4, 2012.
that proposed model Fuzzy Deep Belief Network is superior
to predict heart disease than Deep Belief Network. [8] A. Pethalakshmi and A. Anushya, “Effective Features
Selection via Futuristic Genetic on Heart Data”,
6. ACKNOWLEDGMENTS International Journal of Computational Intelligence and
This project work has been done by S. Saranya and Dr. R. Informatics, Vol. 2, No. 1, pp. 23-27, 2012.
Manavalan, who are Research Scholar (Arignar Anna
[9] Abhishek Taneja, “Heart Disease Prediction System
Government Arts College, Villpuram) and Associate
using Data Mining Techniques”, Oriental Scientific
Professor and Head (Arignar Anna Government Arts College,
Publishing, 2013.
Villpuram) respectively.
[10] R. Chitra and V. Seenivasagam, “Heart Disease
7. REFERENCES Prediction System using Supervised Learning Classifier”,
[1] Chaitrali S. Dangare and Sulabha S. Apte, “A Data Bonfring International Journal of Software Engineering
Mining Approach for Prediction of Heart Disease using and Soft Computing, Vol. 3, No. 1, pp. 1-7, 2013.
Neural Networks”, International Journal of Computer
Engineering and Technology, Vol. 3, No. 3, pp. 23-29, [11] Shamsher Bahadur Patel, Pramod Kumar Yadav and D.P.
2012. Shukla, “Predict the Diagnosis of Heart Disease Patients
using Classification Mining Techniques”, IOSR Journal
[2] M. Anbarasi, E. Anupriya and N. Iyengar, “Enhanced of Agriculture and Veterinary Science, Vol. 4, No. 2, pp.
Prediction of Heart Disease with Feature Subset 61- 64, 2014.
Selection using Genetic Algorithm”, International
Journal of Engineering Science and Technology, Vol. 2, [12] Vikas Chaurasia and Saurabh Pal, “Early Prediction of
No. 10, pp. 5370-5376, 2010. Heart Diseases using Data Mining Techniques”,
Caribbean Journal of Science and Technology, Vol. 1,
[3] Milan Kumari and S. Godara, “Comparative Study of pp. 208-217, 2013.
Data Mining Classification Methods in Cardiovascular
Disease Prediction”, International Journal of Computer [13] Nilakshi P. Waghulde and Nilima P. Patil, “Genetic
Science and Technology, Vol. 2, No. 2, pp. 304-308, Neural Approach for Heart Disease Prediction”,
2011. International Journal of Advanced Computer Research,
Vol. 4, No. 3, pp. 331-338, 2014.

33
 International Journal of Computer Applications (0975 – 8887)
Volume 181 – No. 49, April 2019

[14] R. Chitra and V. Seenivasagam, “Review of Heart [20] K. Rajalakshmi and K. Nirmala, “Heart Disease
Disease Prediction System using Data Mining and Prediction with Map Reduce by using Weighted
Hybrid Intelligent Techniques”, ICTACT Journal on Soft Association Classifier and K-Means”, Indian Journal of
Computing, Vol. 3, No. 4, pp. 605-609, 2013. Science and Technology, Vol. 9, No. 19, pp. 231-237,
2016.
[15] Rupali R. Patil, “Heart Disease Prediction System using
Naive Bayes and Jelinek-Mercer Smoothing”, [21] S. Suganya and P. Tamil Selvi, “A Proficient Heart
International Journal of Advanced Research in Computer Disease Prediction Method using Fuzzy-Cart
and Communication Engineering, Vol. 3, No. 5, pp. 515- Algorithm”, International Journal of Scientific
523, 2014. Engineering and Applied Science, Vol. 2, No. 1, pp. 1-6,
2016.
[16] B. Venkatalakshmi and M.V. Shivsankar, “Heart Disease
Diagnosis using Predictive Data mining”, International [22] Rishabh Wadhawan, “Prediction of Coronary Heart
Journal of Innovative Research in Science, Engineering Disease using Apriori algorithm with Data Mining
and Technology, Vol. 3, No. 3, pp. 223-229, 2014. Classification”, International Journal of Research in
Science and Technology, Vol. 3, No. 1, pp. 1-15, 2018.
[17] Andrea D. Souza, “Heart Disease Prediction using Data
Mining Techniques”, International Journal of Research [23] Herculano-Houzel, Suzana and Lent, Roberto. Isotropic
in Engineering and Science, Vol. 3, No. 3, pp. 74-77, Fractionator: A simple, rapid method for the
2015. Quantification of Total Cell and Neuron Numbers in the
Brain. In: The Journal of Neuroscience. 9 March 2005,
[18] Moloud Adbar et al., “Comparing Performance of Data 25(10): 2518-2521. ISSN: 0270-6474, 2005.
Mining algorithms in Prediction Heart Diseases”,
International Journal of Electrical and Computer [24] Guyon, Isabelle. A Scaling Law for the Validation Set
Engineering, Vol. 5, No. 6, pp. 1569-1576, 2015. Training Set Size Ratio. In: AT&T Bell Laboraties,
1997.
[19] Jaymin Patel, Teja Upadhyay and Samir Patel, “Heart
Disease Prediction using Machine Learning and Data [25] Hinton, G. E. Training products of experts by
Mining Techniques”, International Journal of Computer minimizing contrastive divergence. In: Neural
Science and Communication, Vol. 7, No. 1, pp. 129-137, Computation, 2002, 14(8):1711-1800.
2016.

IJCATM : www.ijcaonline.org 34